Rainer Orth wrote: > Bart Smaalders <bart.smaalders at Sun.COM> writes: > >>>> Apart from the issues mentioned in my last mail (integrating 4.3.3 >>>> instead of the current 4.3.4, defaulting to Studio backend on SPARC), >>>> this seems fine to me. >>> I agreed, defaulting to the Studio backend is a really bad >>> idea. Including the Studio backend is a great idea but the default should >>> be the default backend from upstream. >>> >> The question is one of compatibility to me... the provenance of the >> backend seems irrelevant, but its interface is not. > > Indeed: I fear that e.g. stuff like GCC inline assembler won't work with > the Studio backend, breaking software that would work with FSF GCC out > of the box.
Compatibility is extremely important. This will be our 6th release of gcc for Sparc with the Sun backend as the default. Many applications have been ported to Sparc using these compilers and have achieved high performance. We understand fears regarding regarding compatibility. We take compatibility very seriously, and so far, these fears have not materialized. We will aggressively address any compatibility issues that may arise in the future. Our systems are designed with close cooperation between hardware and compiler engineers. Code generation is a key part of meeting the objectives for the finished product. We make both backends available for users with a preference. The Sun backend provides better code generation for Sparc. The out of the box experience has to be good both in terms of porting ease and application performance. If it is not, we may lose that user and it can be vary hard the get him or her back. GCC with the Sun backend provides the best overall experience on Sparc in terms of open source compatibility, code generation, and support for tools, both Gnu and Sun. It should be the default. Thanks, George
